(palworld) directed by darkside
Viewing sample resized to 55% of original (view original) Loading...
Parent: post #164158 that has a sibling (learn more) show »
Children: 1 child (learn more) show »
Description

Long overdue followup to my previous cattiva gen. Brat status: Tamed.

This took me a long ass time to make, easily the longest I've ever spent on a single image. Even so, it's not exactly what I hoped it would come out to be. Regardless, I liked the idea enough to warrant pushing through, so I'm glad it's done. I think I'll stick to simpler gens for a while now, lol

Gen info

Step 1 - txt2img

masterpiece, best quality, newest,
cattiva, pink body, yellow countershading, blue eyes, (curled tail:1.5), <lora:Cattiva_IL:1>,
head down ass up, (face squish:1.2), (head on ground:1.2), open mouth, tongue, blushing profusely, drooling, male/female, from behind position, sex, size difference, human, interspecies, happy, from side, faceless male, mostly offscreen character, tail grab, hand on head, fucked silly, one eye half-closed, 5 fingers,
cute fangs, paws,
tree, rock, outside, grass,
flong, (twistedscarlett60:0.6), (kekitopu;1.1), syuro, narse, (short stack:1.2), chibi,
Negative prompt: worst quality,
Steps: 35, Sampler: Euler a, Schedule type: SGM Uniform, CFG scale: 5, Seed: 3476130254, Size: 1344x768, Model hash: d2cce406e3, Model: xavier_v10, RNG: CPU, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: base, Lora hashes: "Cattiva_IL: 6041164c1f08", Rescale CFG: 0.7, Version: neo, Module 1: sdxl_vae

I had to reroll this prompt a ton of times until I got something I liked and that wouldn't have required 1000 hours of inpainting. Had to be especially picky because of the outpainting I knew I'd have to do later since my plan was to add text afterward.

Step 2 - img2img (regenning whole image without upscaling)

same prompt as above

Made several img2img passes at high denoise vaules (around 0.5-0.7) to gen new images using roughly the same posing and composition from step 1. Eventually I got one decent enough and selected it as my new base gen.

Step 3 - Manual edits
Removed a TON of artifacting and general AI weirdness. Corrected uneven finger counts on both human and cattiva, repainted eyes, removed extra fang, removed large distracting grass blades at bottom of image

Step 4 - More manual edits
Cropped image to trim off the leftmost 256 pixels.

Step 5 - img2img (outpainting)

masterpiece, best quality, newest,
rock, outside, grass, tree,
flong, (twistedscarlett60:0.6), (kekitopu;1.1), syuro, narse,
Negative prompt: worst quality,
Steps: 35, Sampler: Euler a, Schedule type: SGM Uniform, CFG scale: 5, Seed: 2615398505, Size: 1088x768, Model hash: d2cce406e3, Model: xavier_v10, Denoising strength: 0.98, Init image hash: 7a0f013a674da0554989d02ff0748719, RNG: CPU, Style Selector Enabled: True, Style Selector Randomize: False, Style Selector Style: base, Mask blur: 40, Masked content: fill, Rescale CFG: 0.7, Version: neo, Module 1: sdxl_vae

Used poor man's outpainting script to add 256 pixels of content to the right of the image. This was to extend the background and make extra space for the text

Step 6 - img2img (regenning whole image without upscaling)

Same prompt from step 1

Ran img2img at 0.2 denoise to blend the outpainted portion of the image with the rest of it. This eliminated a vertical seam 256 pixels from the left and blended my edits from step 3 with the AI's rendering

Step 7 - img2img (x2 upscale)

Same prompt from step 1

Ran at 0.22 denoise

Step 8 - Manual edits
Cleaned up some final artifacting

Step 9 - Manual edits
Added text

Blacklisted

    There are no visible comments.